AI applied estimation of learning ρ and λ method
2018/10/01 Yasunori Ushiro (Kanagawa University)
2018/12/03 Addition Learning λ Method
1. Features of learning ρ and λ method
The λ method is a deciphering method in which the ρ method is used for parallel calculation.
The learning λ method is an improvement of the learning ρ method,
and the efficiency of parallel learning is good.
(1) Discovery of learning ability
Discover fixed method for trajectory group of ρ method
Learning ρ method must be solved once, but learning λ method is unnecessary.
(2) Relationship between learning volume and decoding time
The decoding speed of elliptic curve cryptography improves in proportion to the amount of learning.
Compared to ρ method, the learning ρ method is improved by 1,200 times,
the learning λ method is improved by 10,000 times in performance.
(3) Evolution possibility
Application of AI is expected from the use of arbitrary points and fixation of trajectory group
2. Goal of AI application (Change from learning ρ method to learning λ method)
(1) Initial application of AI
1,000,000 times faster than ρ method. Three years later (2021)
It is 34,000 times by the learning ρ method. 30 times faster with AI.
(2) AI intermediate period
100,000,000 times faster than ρ method. 6 years later (2024)
It is 100,000 times by the learning λ method. 1,000 times faster with AI.
Current elliptic curve cryptography is yellow signal.
(3) AI application practical period
1000,000,000,000 times faster than λ method. 10 years later (2028)
It is 1,000,000 times by the learning λ method. 1,000,000 times faster with AI.
Current elliptic curve cryptography is red signal.
3. Destructive power estimation of the United States (NSA)
Prediction from the discovery λ method discovery history and NSA cryptographic capability.
(1) Estimated present (2018)
Learnig ρ:80%, case-(1):50%, case-(2)：30%, case-(3):10%
(2) Estimated five years later (2023)
Learnig ρ:100%, case-(1):95%, case-(2):80%, case-(3):50%